We recently co-authored a Computer Vision whitepaper with the awesome team at Axoltl Innovations. We thought this was a great time to share some background information on our Weapons Detection dataset.

Designing and Capturing a Novel Computer Vision Dataset

I’ve gotten to work on some pretty cool projects in my time here at Rotational Labs, but this is definitely one of the standouts. In early 2025 we were tasked with creating a dataset for firearms detection on live security camera feeds. Designing the dataset meant everything from finding a filming location and planning the camera placement, to curating the appropriate selection of prop weapons and recruiting a team of actors. Capturing the dataset meant orchestrating a very unique film shoot through a hydrid office/industrial setting. Our combined hard work and expertise resulted in something we are really proud of - a powerful domain-specific dataset that is not available anywhere else.

Whitepaper Collaboration

Our friends over at Axoltl Innovations share our passion for all things Computer Vision. We recently had the opportunity to collaborate with them on a whitepaper, in which we explored how their IRIS tool can help expedite the essential (but often very painful) process of annotating vision data. Our weapons detection dataset made for a fascinating case study, and it was impressive to see how quickly we could move from a small set of initial labeled images to a dataset large enough to train baseline comparisons for several popular object detection architectures. Check out the whitepaper for more details, as well as a sneak peak at some of our plans for future collaboration.

The Gritty Details

If you’re considering developing a weapons detection model of your own (or are just a curious fellow CV nerd and want to dive into the details), check out our technical writeup for more information on our novel, proprietary dataset. And as always, reach out to our team with any questions - we’d love to hear from you!

Image generated using Nano Banana